Predicting yellow rust in wheat breeding trials by proximal phenotyping and machine learning

Alexander Koc; Firuz Odilbekov; Marwan Alamrani; Tina Henriksson; Aakash Chawade

doi:10.1186/s13007-022-00868-0

Predicting yellow rust in wheat breeding trials by proximal phenotyping and machine learning

Plant Methods. 2022 Mar 15;18(1):30. doi: 10.1186/s13007-022-00868-0.

Authors

Alexander Koc¹, Firuz Odilbekov^{1

2}, Marwan Alamrani¹, Tina Henriksson², Aakash Chawade³

Affiliations

¹ Department of Plant Breeding, Swedish University of Agricultural Sciences, P.O. Box 190, SE-234 22, Lomma, Sweden.
² Lantmännen Lantbruk, SE-268 81, Svalöv, Sweden.
³ Department of Plant Breeding, Swedish University of Agricultural Sciences, P.O. Box 190, SE-234 22, Lomma, Sweden. aakash.chawade@slu.se.

Abstract

Background: High-throughput plant phenotyping (HTPP) methods have the potential to speed up the crop breeding process through the development of cost-effective, rapid and scalable phenotyping methods amenable to automation. Crop disease resistance breeding stands to benefit from successful implementation of HTPP methods, as bypassing the bottleneck posed by traditional visual phenotyping of disease, enables the screening of larger and more diverse populations for novel sources of resistance. The aim of this study was to use HTPP data obtained through proximal phenotyping to predict yellow rust scores in a large winter wheat field trial.

Results: The results show that 40-42 spectral vegetation indices (SVIs) derived from spectroradiometer data are sufficient to predict yellow rust scores using Random Forest (RF) modelling. The SVIs were selected through RF-based recursive feature elimination (RFE), and the predicted scores in the resulting models had a prediction accuracy of r_s = 0.50-0.61 when measuring the correlation between predicted and observed scores. Some of the most important spectral features for prediction were the Plant Senescence Reflectance Index (PSRI), Photochemical Reflectance Index (PRI), Red-Green Pigment Index (RGI), and Greenness Index (GI).

Conclusions: The proposed HTPP method of combining SVI data from spectral sensors in RF models, has the potential to be deployed in wheat breeding trials to score yellow rust.

Keywords: Disease resistance; Field phenotyping; High-throughput phenotyping; Low-cost phenotyping; Plant breeding; Spectral vegetation index; Winter wheat; Yellow rust.

Abstract

Grants and funding